Ambassador @adaption_ai | AI Engineer | 10Pearls FYP '25. Specialist in LLMs, RAG & LangChain. Building intelligent AI automations & full-stack web apps.
Karachi, Pakistan
- Tweets 3,397
- Following 933
- Followers 1,136
- Likes 12,379
Zero data used to be a hard stop.
Now we just invent the dataset.+17% quality, +19% diversity over top models, and constraints don’t slow it down. Proud @adaption_ai ambassador. This is a real leap for post-training.
The biggest hurdle in frontier AI is curating high-quality, diverse data.
What happens if you need to post-train a model for a new capability, but have zero data?
Today, we release the technical report for Invent a Dataset.
3
144
Tried Invent a Dataset last week.
The new Invent API makes it even smoother: you can build custom, AI-ready datasets in minutes with a few lines of code.
Solid upgrade from @adaption_ai .
#Engineering
Last week we introduced Invent a Dataset. Describe what you need. Hit go.
Today, we’re making it even easier with the Invent API.
A few lines of code. AI ready training datasets in minutes.
102
This is big.
Describe the dataset you need → get a structured, training-ready dataset back.
No existing data required.
Dataset creation just shifted from collection to pure specification.
Impressive work
@adaption_ai
🔥
Introducing Invent a Dataset.
Describe the dataset you need. Get a structured, training-ready dataset back. No existing data needed.
Dataset creation used to start with collection. Now it starts with specification.
54
Ready to move from simple RAG to production-grade AI agents? I'm applying for the AI Engineering Buildcamp taught by @Al_Grigor to master agentic workflows, function calling, and LLM evaluation.
check out this curriculum! Scholarship: forms.gle/nAbwpDKvHAw7cfzh8
2
2
101
🏆 Honored to receive awards in both Part 1 & Part 2 of the @adaption_ai #AutoScientist Challenge, and 🥇 1st Place in Finance!
Thank you @sarahookr , @RossAndAi , @lizasanon , @ecats_ , @IsabelleAtCamp & the entire adaption team.
And thank you for the amazing swag! ❤️🔥
#AI
3
1
8
168
As an Adaption Ambassador, sharing this.
AutoScientist Alignment auto-configures & self-improves RL recipes, less compute, no endless ablations.
Great work
@adaption_ai
Introducing AutoScientist Alignment.
Automatically configure and self-improve your reinforcement learning recipe using AutoScientist.
Save compute and endless ablations. Game the system and allow AutoScientist to do the work.
Now extended to RL recipes.
1
3
206
Just completed the Google AI Professional Certificate, 7 courses, 20+ hands-on AI artifacts, and one vibe-coded app later
Brainstorming, research, writing, content, data analysis, app building, all with AI in the loop.
Verify: coursera.org/verify/professi…
#AI #BuildInPublic
1
104
This is the shift. AutoScientist lets builders own frontier intelligence without needing a full research lab. Own your models. Accelerate your R&D. Stop renting intelligence.
You don't need a research team the size of a lab to own frontier AI. You need AutoScientist.
Own your own intelligence. Accelerate your own R&D.
2
80
Thank you for sharing informative and valuable resources.
Cloud projects are fun until the first bill arrives...
I checked the current terms of four Microsoft resources that can make the early stages much cheaper for DS students, AI builders, and early founders.
Here's what each one actually offers 🧵
1
2
37
Level up your tech skills for FREE with @Microsoft
🎓 $100 Azure Student Credits: azure.microsoft.com/free/stu…
📚 Interactive Hands-On Courses: learn.microsoft.com?wt.mc_id…
Bookmark this for later! 🔖
#TechTwitter #Coding #Azure
1
68
Just submitted AgriIntel-Contextual-Yield-Reasoning-v1 for AutoScientist Part 2 by @adaption_ai ! 🌾🤖
Powered by openai/gpt-oss-120b with multi-step agronomic reasoning.
📈 +175% Quality Score Boost
🎯 Grade E ➡️ Grade C
👇 Links in the comments!
#AI #AutoScientist
Made with AI
1
2
58
We don’t need 70B models for robust Python pipelines. We just need better data. 💡
I built CodeIntel-4B for the AutoScientist Challenge: an edge-deployable reasoning engine that thinks before it codes, powered by @adaption_ai .
Demo below 👇 Links in replies! #AI
2
1
97
📊 Kaggle
The Synthesized Dataset: kaggle.com/datasets/asadulla…
The LoRA Model & Inference: kaggle.com/models/asadullahd…
🤗 Hugging Face
The Synthesized Dataset: huggingface.co/datasets/asad…
The LoRA Model Weights: huggingface.co/asadullahdoga…
38
Just accepted into the @EgoistMachines AI Passport Ideathon!
Excited to leverage my Python & ML background to build secure agent permission flows and a unified memory layer.
Let's win this! @erinmerylstudy @davidobot_
1
38
Try AutoScientist:
adaptionlabs.ai/auto-scienti…
Enterprise data was never just text. It's documents, images, charts, all at once.
AutoScientist is now multimodal, automatically self-improving frontier AI across modalities. You walk away with a frontier model that recognizes the information you care about.
1
30
Part 2 of my AutoScientist Challenge is live!
Meet MarketIntel-Macro-Reasoning-Mistral-7B 🧠📊
📈 Financial domain win-rate: 13% ➔ 87%
📈 General win-rate: 45% ➔ 56% (Zero catastrophic forgetting!)
Built on @adaption_ai . Weights open-source! 👇
🔗huggingface.co/asadullahdoga…
1
2
32
Join part 2 of the AutoScientist Challenge 💪
Your idea could be next → adaptionlabs.ai/blog/autosci…
The next breakthrough AI model could come from anyone.
Someone with a unique perspective. A problem worth solving. A domain that needs better intelligence.
$60,000 in prizes. Your idea could be next.
4
109
Most HR AI models predict a binary flight-risk score.
I fine-tuned CorpIntel-HR-Agent (Llama-3.2-3B) to output full reasoning + retention plans from raw telemetry.
Built for @adaption_ai AutoScientist Challenge
🤗 huggingface.co/asadullahdoga…
Made with AI
1
1
36
📊 Datasets:
HF: huggingface.co/datasets/asad…
Kaggle: kaggle.com/datasets/asadulla…
1
1
19